Wallaby counts - Department of Conservation

Transcription

Animal pests: distance sampling for
Bennett’s wallaby
Version 1.0
This specification was prepared by David M. Forsyth and Kennedy Lange in 2012.
Contents
Synopsis .......................................................................................................................................... 2
Assumptions .................................................................................................................................... 3
Advantages...................................................................................................................................... 4
Disadvantages ................................................................................................................................. 4
Suitability for inventory ..................................................................................................................... 4
Suitability for monitoring................................................................................................................... 5
Skills ................................................................................................................................................ 5
Resources ....................................................................................................................................... 6
Minimum attributes .......................................................................................................................... 6
Data storage .................................................................................................................................... 7
Analysis, interpretation and reporting ............................................................................................... 7
Case study A ................................................................................................................................... 8
Full details of technique and best practice ....................................................................................... 9
References and further reading ......................................................................................................26
Appendix A .....................................................................................................................................28
Disclaimer
This document contains supporting material for the Inventory and Monitoring Toolbox, which
contains DOC’s biodiversity inventory and monitoring standards. It is being made available
to external groups and organisations to demonstrate current departmental best practice.
DOC has used its best endeavours to ensure the accuracy of the information at the date of
publication. As these standards have been prepared for the use of DOC staff, other users
may require authorisation or caveats may apply. Any use by members of the public is at
their own risk and DOC disclaims any liability that may arise from its use. For further
information, please email [email protected]
Inventory and monitoring toolbox: animal pests
DOCDM-787258
DOCDM-787258 Animal pests: distance sampling for Bennett’s wallaby v1.0
2
Synopsis
Figure 1. Bennett’s wallaby (photo: G. Coulson).
The suitability of distance sampling to estimate the density and abundance of Bennett’s wallaby
(Macropus rufogriseus) in a 3875 ha area of montane grassland habitat in the Hunters Hills was
assessed in February 2010. Distance sampling showed promise for estimating the density and
abundance of Bennett’s wallaby. The way that distance sampling will be employed to estimate the
density and abundance of Bennett’s wallaby will depend on circumstance (e.g. habitat, history of
hunting/control, population density), but we recommend the following key steps:
1. Survey objective(s) and area
Define the survey objective(s) and the study area. Is distance sampling the best of the
available methods to address the survey objective(s) given the study area’s size, terrain,
and vegetation, particularly in light of the required resources? If yes, proceed to the survey
design and planning stage.
2. Survey design and planning
The best approach is to systematically sample the study area with transects that sample
across the grain of the habitat (i.e. across gullies rather than down or up). The study area
may be stratified on the basis of habitat (e.g. grassland and forest) or expected densities
(e.g. as a consequence of control history). A minimum of 15–20 transects is recommended,
and consideration should be given to zigzag transects to minimise time spent walking
between transects. For small study areas, fewer transects can be sampled multiple times. A
minimum of 60–80 observations are required to robustly estimate density using distance
sampling, so consideration needs to be given to how much survey effort (i.e. number of
transects and number of times each transect is walked) is required to achieve this. It may be
prudent to walk a sub-sample of transects once to estimate what resources will be required
for the survey in terms of people, equipment and transport, etc.
3. Observer training
Careful training of observers is required, and resources need to be allocated to this task.
3
Observers must understand how to walk and search transects, and how to use equipment
(e.g. a laser range finder and sighting compass) correctly.
4. Data collection
Data should be collected as accurately as possible and entered into a standardised field
sheet. Data should be checked by the field team leader after the first transects have been
completed and again at the end of the field trip.
5. Data entry and storage
Field sheets should be scanned and saved as *.pdf documents, and electronic and hard
copies stored and archived. The data should be entered into a spreadsheet or database and
archived, and imported into the latest version of DISTANCE (Thomas et al. 2009) for
analysis.
6. Analysis
The analyst(s) should familiarise themselves with how other macropod distance sampling
studies have been analysed. First, exploratory data analysis is conducted to understand the
data and identify problems. Particular problems in wallaby surveys would be too few
detections close to the transect line (most usually caused by animals moving away before
the observer detects them) or, as occurred in our study, a ‘spike’ of observations at zero
(usually caused by observers rounding small angles down to zero). The second stage of
analysis is model selection. Truncation (left and/or right) of data and definition of
intervals/cutpoints (for histograms, goodness-of-fit and other diagnostic tests) are conducted
at this stage. The third and final stage of analysis is inference from the best model(s).
7. Reporting
The final report should include sufficient information so that your study could be duplicated
by another person without your assistance. It would therefore be useful to report the
electronic locations of the scanned field sheets, stored data and DISTANCE project. The
report should also include the study objectives and design, the observers used and any
training given, and a detailed summary of the three stages of analysis (i.e. exploratory data
analysis, model selection and inference from the best model(s)).
Assumptions






All animals on the transect line are detected (p(0) = 1), although violation of this assumption may
be accommodated.
Animals do not move prior to detection. A snapshot is usually obtained—all animals are
detected as if stationary.
Distances from the transect line to animals are accurately measured.
Individuals, or clusters of individuals, are detected independently of other such sample units.
Transects are distributed over the area of interest according to a probability-based sampling
design (simple random, systematic, stratified, etc.).
The animal population remains constant throughout the survey period.
4
Advantages






If the assumptions hold, distance sampling can provide robust estimation of abundance more
cheaply than mark–recapture/resight methods.
Not all animals within the study area need be counted.
Compared to mark–recapture methods, the modelling component is relatively straightforward
(largely a software issue).
Providing the first assumption holds, the potential for bias introduced by pooling data from
observations with different detection probabilities is significantly reduced (i.e. they are ‘pooling
robust’). This is a very powerful feature of distance sampling.
Unbiased population estimates can be compared across time and space.
Analysis of distance data has been dramatically simplified by the software DISTANCE (Thomas
et al. 2009).
Disadvantages






Violation of critical assumptions may cause serious errors in density and abundance estimates.
These assumptions are usually violated to some degree in most animal surveys using this
method: the question is to what extent.
Application of the method in some habitats (e.g. thick forest) can be problematic.
Sampling design must be considered on a case-by-case basis given the topography and
vegetation in the area of interest.
The minimum number of detections required to model the detection function is relatively large.
As a rule of thumb, 60 detections are required as a minimum for distance sampling surveys, and
estimates tend to improve with more detections. Distance sampling is therefore unlikely to be
suitable for low-density populations, particularly if it is necessary to stratify data to account for
habitat and observer differences. Pooling data may alleviate this potential problem in some
circumstances.
Observer training is necessary to ensure that the key assumptions of distance sampling are, as
far as is possible, met. It should be recognised that not all field staff may be competent to
conduct distance sampling. Observer training is therefore a critical component of any distance
sampling programme.
The cost and effort required to obtain data suitable for unbiased and precise population
estimates may exceed available funding.
Suitability for inventory
The expense of distance sampling is not usually justified for inventory.


Specialist skills for design and analysis are also likely to be in short supply and expensive.
Costs (labour and money) can be large and the results obtained often beyond that required for
simple inventory.
5
Suitability for monitoring



If the critical assumptions listed above can be met (or at least not grossly violated) and sufficient
resources are available, distance sampling can provide robust and unbiased estimates of
density and abundance.
Comparison of density/abundance estimates over time and across space is possible and this is
an advantage for monitoring programmes in which the primary objective is to estimate absolute
density and/or population size. However, the relative merits of absolute density estimates and
indices for describing trend depend on their relative sampling variance (including the effect of
operational overheads on sample size), the level of uncontrollable variation in detectability, and
the intended audience.
Distance sampling is less suitable for monitoring low density animal populations simply because
obtaining the minimum number of observations (60) is likely to be prohibitive in terms of time,
labour and cost. However, if the detection function is known from a previous survey (e.g. a precontrol survey when the population was at a higher density) then program DISTANCE can be
used to estimate density using fewer than 60 observations (this application is not discussed
further here).
Skills
Those responsible for survey design must:


Be familiar with the relevant design issues pertinent to the use of distance sampling methods for
animal populations. Buckland et al. (2001, 2004) devote many chapters to survey design and
field methods. These issues include the critical assumptions and their impact on appropriate
sampling design, definition of the sampling frame and sampling unit, the number of transects
(and their length) and distribution of transects over the sampling frame.
Have an understanding of the target species’ spatial distribution (e.g. clumped or territorial) and
potential for stratification. This understanding is also extremely useful and can markedly improve
the precision of abundance estimates. A pilot study is strongly recommended (Thompson et al.
1998) because it will provide useful information on the precision resulting from a given level of
labour and cost. It will also provide an idea of the expected encounter rate and detection
function from which the transect length required to reach predetermined levels of precision can
be estimated.
Field observers must be:



Familiar with the target species (identification, behaviours, etc.)
Able to consistently follow the designated sampling design and make accurate distance and
bearing measurements
Able to identify violations of assumptions and understand the consequences for density and
abundance estimates (see Borchers et al. 2002, p. 160)
Those responsible for analysis must have:

Specialist statistical skills and familiarity with the program DISTANCE.
6
Resources
Distance sampling is usually more expensive than obtaining indices of relative abundance despite
the relatively small amount of additional information collected in the field. This is particularly true if
the species of interest is at low to moderate density, as more intensive sampling and/or more
transects are required for reasonable precision. Forethought is required when designing the
sampling programme to ensure the critical assumptions underlying distance sampling are met and
sufficient data are collected. Additional sampling infrastructure is often required to ensure that
transects are sufficiently well defined to minimise measurement error (e.g. marker poles, tags, etc.).
All sampling programmes utilising distance sampling will require a significant training component to
ensure observer competence in the data collection methods, equipment use and species
identification. Consider testing of potential observers’ visual acuity—use only those who reach a set
standard. If training and checks are not done, an additional cost will be the bias introduced by the
violation of underlying assumptions.
This method requires the following resources:









Suitably trained people
Maps of the study area and transects to be sampled
Marked transects
GPS to assist with navigating to/from and along transects
Range finder to measure distance
Compass to measure bearing
Data sheets and notebook
Suitable means of moving between plots (e.g. walk, drive, helicopter)
Appropriate communication, safety and first-aid equipment
Minimum attributes
Consistent measurement and recording of these attributes is critical for the implementation of the
method. Other attributes may be optional depending on your objective. For more information refer
to ‘Full details of technique and best practice’, Buckland et al. (2001) and Thomas et al. (2009).
DOC staff must complete a ‘Standard inventory and monitoring project plan’ (docdm-146272).
In the field, record data on the field sheet, preferably printed on waterproof paper.
Minimum attributes to record:




Metadata, including observer’s name and contact details, date of survey, time over which survey
conducted (start/finish times) and relevant weather details
Location (eastings and northings and/or polygons) of survey area, sample area, lines or points
and strata (if required)
Habitat variables associated with line/point and stratum
Line length and sample effort (number of times line walked or point visited)




7
Number of individuals (or group of individuals, termed ‘cluster’) of the target species seen from
the transect line
Distance (to nearest metre) or distance interval from the transect line to each individual or
cluster
Bearing (to nearest degree) from the transect line to each individual or cluster
Covariates that may help explain density (e.g. treatment, non-treatment, vegetation type)
Data storage





Copies of completed field sheets should be forwarded to the survey administrator, photocopied
or scanned, then entered into an Excel spreadsheet as soon as possible after returning from the
field. The key steps here are data entry, storage and maintenance for later analysis, coupled
with copying and data backup for security. If data storage is designed well at the outset, it will
make the job of analysis and interpretation much easier. Before storing data, check for missing
information and errors, and ensure metadata are recorded.
Storage tools can be either manual or electronic systems (or both, preferably). They will usually
be summary sheets, other physical filing systems, or electronic spreadsheets and databases.
Use appropriate file formats such as .xls, .txt, .dbf or specific analysis software formats. Copy
and/or backup all data, whether electronic, data sheets, metadata or site access descriptions,
preferably offline if the primary storage location is part of a networked system. Store copies at a
separate location for security purposes.
Completed data sheets and/or field note books can be scanned and saved as a pdf with a selfexplanatory file name and stored, along with the entered data, in multiple physical locations to
guard against potential loss of data from server failure and natural disasters.
Our preference is to have one data sheet containing all the information collected in the field (and
also including transect beginning and end eastings and northings) and have a separate figure
(map) showing the layout of transects, but separate worksheets or files could contain that
information and any other explanatory material.
Not all of the data entered in the spreadsheet is imported into the DISTANCE software, and our
preference is to create another worksheet (‘Distance data’) containing those data. That
worksheet is then saved as a text file (i.e. *.txt) and imported into DISTANCE using the Import
Data Wizard (Thomas et al. 2009). The imported data and associated analyses and output can
be stored as ‘Distance Projects’ (Thomas et al. 2009).
Analysis, interpretation and reporting
The user of DISTANCE is directed to the software’s user guide1 (Thomas et al. 2009) where the
more complex mechanics of this analysis programme are explained in detail. Buckland et al. (2001,
pp. 48–50) suggest a useful strategy for the analysis of data sets and this is summarised below.
1
http://www.ruwpa.st-and.ac.uk/distance/
8
Exploratory phase
During this initial phase, the analyst should critically examine the data collected. Take particular
care to code effort (the number of times a sample unit has been surveyed) correctly to avoid
pseudoreplication (Thomas et al. 2009). Identify and correct any data entry errors and other
anomalies. Plot the data as histograms in a variety of groupings so the data can be examined in
detail. These histograms can be used to identify outliers, as well as potential violation of
assumptions caused by the presence of ‘heaping’ and ‘spiking’ (distances rounded to certain
distances) and evasive movement (see Westbrooke et al. 2003). Data can be grouped to reduce
these impacts. Buckland et al. (2001, p. 151) recommend truncation of larger distances to reduce
the impact of outliers and improve model fit.
Model selection
Once a dataset has been prepared, several robust models describing the detection function should
be considered. Program DISTANCE provides several useful models or ‘key functions’ and
associated adjustment terms (or ‘series expansions’) used to ‘shape’ detection functions to fit the
data. Likelihood ratio tests, goodness-of-fit tests and Akaike’s information criterion (AIC) are all
available as aids to objective model selection. Often the need for additional exploratory work
becomes apparent at this point. Changes, such as altering truncation point, regrouping distance
intervals or pooling data across surveys, might improve the fit of one or more of the candidate
models. It is common to find there are several competing models, some of which appear to perform
poorly relative to other models (and can be discarded) and others that perform equally well. Model
averaging can then be used to account for model selection uncertainty (Burnham & Anderson
2002).
Final analysis and inference
Selection of the model(s) believed to be the best fit for the data can now occur. Once a single
model (or subset of models) has been selected, estimates of density and precision of these
estimates can be made (along with relevant tables and graphs) and then discussed in relation to
any perceived failures of critical assumptions. Improved variance estimates can be obtained using
bootstrapping routines and, if necessary, including a variance component to reflect model-selection
uncertainty.
Case study A
There is no case study available for this method.
9
Full details of technique and best practice
Objective
The objective of the monitoring must be explicitly stated. The objective of our study was:

To estimate the density and abundance of wallabies in a 3875 ha area of the Hunters Hills prior
to control being undertaken
Sampling design
The boundary of the study area needs to be delineated on a map (e.g. with a GIS). The boundary of
our study area was defined by Raukapuka Area Office staff based on local knowledge and the
intention to conduct wallaby control there during winter 2010 (Fig. 2).
The sampling design needs to be determined, and many options are available (see Buckland et al.
2001, 2004). The sampling design used in our study consisted of 16 line transects arranged parallel
(i.e. systematically) to sample across the grain of the terrain (i.e. from ridge down across the gully to
the opposite ridge) (Fig. 2). Sixteen was the practical minimum number that could be walked once
by the field staff in the available time to ensure a minimum 60 observations of Bennett’s wallaby.
The length of the transects (estimated with a GIS) varied from 1202 m to 3594 m, and the surface
area of the study area was 3875 ha.
Prior to going into the field, staff were briefed in the office on the study’s objective and design, the
theory of distance sampling and the need to accurately record sighting distances and bearings, and
were shown how to use the laser range-finders (Leupold® TX™-II 6×32 Digital Range Finder) and
sighting compass (Suunto KB-series).
The transects were not marked. Rather, field observers were provided with a map and a GPS unit
(Garmin CSx 60) with the transects loaded, and field sheets printed on waterproof paper. Observers
were also provided with a compass and a sheet with the end coordinates and compass bearings for
each transect.
Staff who had not conducted distance sampling before were accompanied on their first transect by
an experienced observer. Observers therefore walked transects either alone or in pairs, following
the transect as shown on the map and GPS as closely as possible.
Staff were assigned transects in a way that maximised time on transects and minimised time spent
travelling between transects. Four-wheel drives, quad bikes and a helicopter were used to ferry
observers to/from the ends of transects. Transects were walked once at 1–2 km-1 (i.e. a slow
walking speed) in good visibility throughout the day. No sampling was undertaken in fog or rain, or
at night. Unsafe terrain was occasionally encountered: observers were instructed to walk around
the terrain and to return to the transect line as soon as possible. The most practical method for
walking transects in our study area (Fig. 3) was to use the map and compass to identify the line to
be walked and select a feature (e.g. rock or tree) ahead along the line to walk towards. The
10
observer then walked towards the landmark, scanning forward left and right from the transect line,
but with most search effort focused on and around the transect line. When one or more wallabies
was sighted, the distance from the observer to the ‘cluster’ (defined as a group of wallabies; the
minimum cluster size is one) was measured with the range finder and the bearing was measured
with the sighting compass. Any other biota or items of interest could also be recorded on the data
sheet (e.g. chamois Rupicapra rupicapra were seen along one transect).
Figure 2. Boundaries of the study area showing the location of the 16 numbered line transects that were each
walked once in February 2010.
11
Figure 3. Typical view along a transect in our study area (photo: D.M. Forsyth).
Data collection
The field sheet used to collect the data required for distance sampling (Fig. 4) was printed on
waterproof paper. You may need to record additional data depending upon the objectives of your
study.
The columns were as follows:




‘Cluster size (# animals)’ is the number of animals in the group (minimum is 1).
‘Distance (m)’ is the (radial) distance in metres from the observer (standing on the transect line)
to the cluster.
‘Angle (0–90°)’ is the bearing from the observer to the cluster.
‘Habitat (Tussock, Scrub or Bush)’—since wallabies would be expected to be more visible in
tussock compared to scrub and bush, each cluster was assigned to one of these vegetation
types.
The field work was conducted in 2 days (9 and 12 February 2010) by five people (i.e. a total of 10
staff field-days). Visibility was excellent and wind was classified as either calm or light on all 16
transects. A total of 85 clusters was sighted on the 16 transects, which is greater than the minimum
60–80 recommended for robust estimation of density using distance sampling by Buckland et al.
(2001).
Data storage
Following completion of the survey, data sheets were photocopied and scanned (as a pdf) before
data were entered into the worksheet ‘All data’ in the Excel spreadsheet
‘Wallaby_data_DISTANCE_28Feb2010.xls’ (Fig. 5).
Raukapuka Bennett’s wallaby DISTANCE data
Date:
Cloud cover (%):
Transect:
Start coordinates:
End coordinates:
Cluster size
(# animals)
Bearing:
Distance
(m)
Angle
(0–90°)
Observer(s):
Wind: Calm/Light/Moderate/Gale
Length:
Time started transect:
Time finished transect:
Habitat (Tussock, Scrub
or Bush)
Figure 4. Field sheet used to collect distance sampling data.
Notes
12
13
Figure 5. Excel worksheet in which all the information on the field data sheet (see Fig. 4) has been entered.
Figure 6. Excel worksheet containing the data to be imported into DISTANCE. The worksheet is saved as a text file
(i.e. *.txt) for import.
14
Analysis
The data to be imported into DISTANCE were copied and pasted from the ‘All data’ spreadsheet
into the ‘DISTANCE data’ worksheet (Fig. 6), and the latter worksheet was saved as a text file
‘Wallaby_data_DISTANCE_28Feb2010.txt’ and imported into the freeware DISTANCE 6.0 version
2 (Thomas et al. 2009).2 The data following import into DISTANCE are shown in Fig. 7, and the
column variables in Fig. 7 are defined in Table 1.
Figure 7. Data from Figure 6 after importing into DISTANCE 6.0 version 2.
2
Available at http://www.ruwpa.st-and.ac.uk/distance/
15
Table 1. Definition of column variables in Figures 6 and 7.
Spreadsheet column
Corresponding column in DISTANCE
Explanation
-
Study area: Label
Contains information that applies to
the whole study.
A
Region: Label
Name of study area or stratum.
B
Region: Area
Area of region (ha) used to calculate
density and abundance.
C
Line transect: Label
Transect number: in this example
there were 16 (Fig. 2), with only
three visible on the Fig. 7 screenshot
(i.e. transects 1, 2 and 3).
D
Line transect: Line length
The length of each of the 16
transects in metres (estimated with a
Geographic Information System).
E
Observation: Radial distance
Radial distance in metres from the
transect to the cluster of animals.
F
Observation: Angle
Angle in degrees from the transect to
the cluster of animals. The angle can
range from 0 to 90.
G
Observation: Cluster size
Number of wallabies in the cluster.
The minimum cluster size is 1.
Exploratory analyses
We first conducted exploratory data analyses to examine the distribution of observations as a
function of distance from the transect line. We did this by defining a set as ‘Exploratory data
visualisation’ in the Analysis browser. All distances reported henceforth are perpendicular distances
calculated in DISTANCE. The most distant observation was 174.3 m from the line.
We first plotted the distribution of sightings in 10 m intervals from 0–200 m (Fig. 8). That histogram
revealed that there was strong evidence of ‘spiking’ in the first distance class (0–10 m) and a very
long ‘tail’ (i.e. there were many observations beyond about 80 m that would distort the detection
function model). The histogram is similar to that reported for eastern grey kangaroos (Macropus
giganteus) on Rotamah Island (Coulson & Raines 1985), although spiking is more pronounced in
Fig. 8 and is usually caused by observers rounding bearings close to the transect line to zero
(Buckland et al. 2001).
16
Figure 8. Histogram of the number of clusters of Bennett’s wallaby observed in a 3875 ha area of the Hunters Hills
as a function of perpendicular distance from the transect line. These data indicate spiking in the 0–10 m distance
class, and a very long tail requiring truncation at about 80 m.
Model selection
The first step in model selection involves selecting a suitable truncation distance w for the data. The
truncation of larger distances is recommended so as to reduce the impact of outliers and improve
model fit (Buckland et al. 2001, p. 151). Inspection of Fig. 8 reveals that little additional information
(in terms of estimating a detection function) is contained in the 14 sightings beyond 80 m, so we
chose w = 80 m. Although a greater percentage of sightings were truncated (17%) compared to the
standard recommendation (5%) of Buckland et al. (2001), up to 17% of data were truncated in
ground surveys of multiple macropod species by Clancy et al. (1997), so this is not without
precedent. Moreover, this highlights the need to carefully consider data on its own merits.
Because of the spiking evident around the transect line (Fig. 8), we next grouped our data so as to
improve estimates of density and abundance. We chose a relatively wide first interval to ensure that
as few as possible detections were erroneously allocated to the first interval through measurement
error and, in particular, through rounding a small sighting angle to zero (Buckland et al. 2001, pp.
109–110). We explored the consequences of other truncation distances (e.g. 110 m), numbers of
intervals and interval widths manually in the Data Filter option of the ‘Analyses’ tab in concert with
the following model options (Thomas et al. 2010): uniform key with cosine adjustments, half-normal
key with cosine adjustments, half-normal key with Hermite polynomial adjustments. We used the
automated selection of adjustment terms function, with sequential one-step selection up to a
maximum of five terms based on AICc. We also evaluated the effects of estimating cluster size as
the mean of observed clusters or from size-biased regression on density estimates. We assessed
the fit of the various models to our data with histograms and goodness-of-fit tests along with our
biological understanding of wallabies in the study area. (Note that we could not use quantile–
quantile plots and Cramér–von Mises tests as diagnostic tools for these analyses because our data
were analysed as grouped rather than as exact distances because of the spiking on the transect
line.)
17
Of the four key-adjustment models considered, the hazard-rate model fitted an implausible shape to
the detection function (i.e. no shoulder) and is not considered further here, and neither the cosine
nor Hermite polynomial adjustment was required for the half-normal model. Hence, only the halfnormal model (with no adjustments) and the uniform key with cosine adjustments are considered
further. Diagnostic output from DISTANCE (histograms with χ2 goodness-of-fit tests) is shown below
for the two models (Figs 9 and 10), and suggest that both models provided reasonable fits to the
data. The AICc values for each model were also very similar (half-normal = 217.42; uniform+cosine
= 217.92), indicating that both models should be considered in the third and final stage of analysis.
Note that the model selection phase involved considering other cutpoints, numbers of intervals,
values of w, and estimation of group size using size-bias regression. The results of each of those
models is not reported here for brevity (but are stored in the DISTANCE project), but in summary
the estimates of density and abundance did not change greatly as these variables were changed.
The models reported here are the most sensible of those evaluated. There were too few
observations to estimate detection functions for each of the three habitat types (tussock, scrub and
bush) separately.
Final analysis and inference
The third and final phase of analysis is selection of the best models and reporting estimates of
density and abundance and their associated uncertainties. The two models in Figs 9 and 10 are
virtually indistinguishable in terms of model diagnostics, and their density and abundance estimates
are similar (Fig. 11).
Because the estimates of density were so similar for the two models, we selected the half-normal
model (upper row in Fig. 11) as the model to make our inferences from. Note that this is a
conservative approach, as model-averaging would in this example decrease the confidence interval
widths. The density estimates from DISTANCE are shown in Fig. 12. The output shows effort (the
total length of transect walked) as 37 380 m, that there were 16 transects, that the maximum width
(w) was 80 m, that there was no left-truncation (0 m) and that 71 observations were used in the
analyses (recall that 14 were removed following truncation at w = 80 m). The mean estimated
density was 0.2544 wallabies ha-1, with a 95% confidence interval (CI) of 0.15 to 0.43 wallabies
ha-1, which translated to 986 wallabies (95% CI; 582–1669) in the survey area. The lower table in
Fig. 12 indicates that the primary cause of uncertainty in the estimates of density and abundance
was the estimated encounter rate (i.e. transect-to-transect variation in the number of wallabies
observed per unit length of transect) rather than the estimated detection function or estimated
cluster size.
The next page of DISTANCE output summarises encounter rates (Fig. 13). The variable ‘n/L’ is the
number of wallaby clusters encountered per m of transect (i.e. 0.0019; 95% CI, 0.0012–0.00311):
these are more usefully reported per km of transect (1.9; 95% CI, 1.2–3.1), and would be helpful for
planning further distance sampling monitoring programmes for this species in this type of habitat.
The next page of DISTANCE output summarises detection probabilities (Fig. 14). Perhaps the most
useful information here is the estimated Effective Strip Width (ESW), in this case 42.1 m (95% CI,
18
34.7–51.0 m). If all objects were detected out to that distance on either side of the transect (and
none beyond), then the expected number of objects detected would be the same as for the actual
survey (Buckland et al. 2001, p. 53).
The next page of DISTANCE output is the estimated cluster size (Fig. 15). We set this option to
‘Mean observed cluster size’, and this was 1.12 wallabies (95% CI, 1.03–1.24). Note that we also
investigated the effect of estimating cluster size from a size-bias regression, but that had little effect
on the estimated densities.
19
Figure 9. Diagnostic output for the half-normal model from DISTANCE. The upper panel is the histogram of
2
detection probability for the specified intervals, and the lower panel shows the χ goodness-of-fit tests. Note that as
these data were analysed as grouped (cf. exact) distances, other model diagnostic options were unavailable.
20
Figure 10. Diagnostic output for the uniform key with cosine adjustment model from DISTANCE. The upper panel
2
is the histogram of detection probability for the specified intervals, and the lower panel shows the χ goodness-of-fit
tests. Note that as these data were analysed as grouped (cf. exact) distances, other model diagnostic options were
unavailable.
Figure 11. Model selection summary output from DISTANCE for the two best models estimating density and
abundance of Bennett’s wallaby in a 3875 ha area of the Hunters Hills.
Figure 12. Density and abundance estimates for the half-normal key model from DISTANCE.
Figure 13. Encounter rate estimates for the half-normal key model from DISTANCE.
21
22
Figure 14. Detection probability estimates for the half-normal key model from DISTANCE.
Figure 15. Cluster size estimates from DISTANCE.
Figure 16. Estimated densities of wallaby clusters (DS) and wallabies (D), and estimated number of wallabies (N),
in the 3875 ha study area from DISTANCE.
The final page of output from DISTANCE summarises the estimated densities and abundances of
wallabies (Fig. 16). The information of most interest is the estimated density (D) and abundance
(N), and their 95% CIs. The ‘%CV’ is the coefficient of variation (%CV = [standard
error/estimate]*100%), which is a measure of uncertainty in the density and abundance estimates
(in this case 25.8%).
Limitations and points to consider
The two main limitations of our study were: (a) the relatively small number of observations for the
effort expended (85 observations for 37 km of transect walked in 10 staff-days, excluding planning
and training time), and (b) the evidence of spiking near the transect line. The first limitation is a
consequence of the low density of wallabies in the study area and hence nothing can be done
23
except to walk transects more than once and/or to add additional transects (i.e. greater survey
effort). Hence, it is likely that the resources required to conduct distance sampling will often exceed
those available (or required) for other methods (Warburton & Frampton 1993).
The second limitation, spiking of observations near the transect line (Fig. 8), was most likely caused
by rounding bearings close to the transect line to zero. We believe that permanent marking of
transects (e.g. with marker poles) would likely reduce this problem. It was not feasible to mark
transects before we walked them in our study: doing so would have doubled the field work.
However, given that the study described here has shown promise, we believe that future surveys
should mark the transects prior to sampling. The transects should be marked such that the observer
can always see one marker ahead and one marker behind: the observer need only walk in a
straight line from marker to marker. We overcame the problem of spiking in our analyses by
assigning a wider first interval (0–20 m) relative to the other intervals (all 10 m) (Figs 9 and 10) and
by analysing the data as ‘grouped distances’ rather than ‘exact distances’: neither solution is ‘ideal’,
but data collected in the real world are seldom ideal! It is hoped that data collected from
permanently marked transects would not be subject to spiking and that the data could be analysed
as exact distances.
This is the first study to attempt to estimate the absolute density of Bennett’s wallaby in South
Canterbury using distance sampling. Warburton & Frampton (1993) reported a density of
3 wallabies ha-1 in the Tasman Smith Scenic Reserve based on a carcass search following a
poisoning operation, and this is likely to be the higher end of the density continuum (Warburton
2005). Our study area was at a higher elevation than the Tasman Smith Scenic Reserve and
consisted primarily of sub-alpine grasslands (Figs 2 and 3) with only occasional patches of scrub
along lower-elevation watercourses. Hence, we would expect wallaby densities to be lower in our
study area relative to many other parts of their range on public conservation estate.
We chose to systematically sample the study area using parallel, equally spaced transects that
sampled across the grain of each catchment (Fig. 2). However, this design means that data are not
collected while moving between transects. A potential modification to our design that would
maximise time spent walking transects would be to use a zigzag design (Buckland et al. 2001;
Thomas et al. 2010).
With more observations it may have been possible to fit different detection functions to each of the
three habitats (or to pooled subsets of habitat). It is highly likely that the detection function for
wallabies in tussock would differ from that in scrub or bush, but a minimum of 60–80 observations
would be required in each habitat before detection functions could be robustly modelled separately
for each habitat. Future studies should at least consider this possibility in the design and analysis
phases. A related issue is that of stratifying study areas on the basis of expected wallaby density
and/or habitat. For example, if one wanted to estimate population size then it is usually most
efficient to conduct more effort in higher density strata (Buckland et al. 2001). However, in some
situations it may be desirable to allocate effort to ensure that the minimum 60 observations are
obtained in all strata. During our study we saw more wallabies, and more wallaby sign, in the lowerelevation scrub and tall grassland habitats in gullies. Depending on the study objective(s), it might
24
be efficient to stratify the study area into vegetation types and allocate effort (i.e. transect length per
unit area) according to vegetation type.
We hope that our description of analyses highlighted the multiple options available when analysing
data from line transects when using the DISTANCE analysis software. Indeed, the detailed options
and recommendations for analyses provided by Buckland et al. (2001) and Thomas et al. (2009)
can sometimes seem overwhelming. We believe that the primary objective should be to obtain
defensible density and abundance estimates that are fit for your purpose, and at all times the
analyst should be thinking about the likely behaviour and ecology of the study animal in the study
area in relation to the way the line transect sampling was conducted. For example, we truncated our
data at 80 m rather than using the recommendation by Buckland et al. (2001) to truncate at the
distance that removes 5% of observations because the rolling terrain in parts of our study area
meant that in some situations wallabies were seen very far away (e.g. when traversing a ridge). The
recommendations provided by Buckland et al. (2001) and Thomas et al. (2009) are based on
analyses of many taxa and study areas, but that does not mean they will always be appropriate to
your situation.
More generally, our study highlights the need for observer training. Our study included the bare
minimum training of observers (all but two had not conducted distance sampling previously) and the
concept of searching most intensively on and near the transect line rather than simply counting the
most wallabies on the transect is not something that can be grasped by all field staff. Similarly, not
all staff can use range finders and sighting compasses in the manner required for distance
sampling. We therefore recommend that more intensive training be given and that staff can only be
observers once a required level of competency has been achieved (e.g. by a practical assessment
or periodic checks).
The approach described here shows promise for estimating the density and abundance of Bennett’s
wallaby. The way that distance sampling will be employed to estimate the density and abundance of
Bennett’s wallaby will depend on circumstance (e.g. habitat, history of harvesting/control, population
density), but the following three key references should be consulted: Buckland et al. (2001),
Thomas et al. (2009) and Thomas et al. (2010). Based on our experiences, we recommend the
following key steps:
1. Survey objective(s) and area
The objectives of the survey need to be carefully considered and explicitly defined, and this
includes delimiting the boundaries of the survey area in space and time. Is distance
sampling the best of the available methods to address the survey objective(s), particularly in
light of the required resources? If yes, proceed to the survey design and planning stage.
2. Survey design and planning
Detailed information on survey design is available in Buckland et al. (2001) and Thomas et
al. (2009, 2010). In general, the best approach is to systematically sample the study area
with transects that sample across the grain of the habitat (i.e. across gullies rather than
down or up). The study area may be stratified on the basis of habitat (e.g. grassland and
forest) or expected densities (e.g. as a consequence of control history). A minimum of 15–
20 transects is recommended, and consideration should be given to zigzag transects to
25
minimise time spent walking between transects. A minimum of 60–80 observations are
required to robustly estimate density and abundance using distance sampling, so
consideration needs to be given to how much survey effort (i.e. number of transects and
number of times each transect is walked) is required to achieve this. It may be prudent to
walk a sub-sample of transects once to estimate what resources will be required for the
survey in terms of people, equipment and transport, etc.
3. Observer training
Careful training of observers is required, and resources need to be allocated to this task.
Observers must understand how to walk and search transects, and how to use equipment
(e.g. a laser range finder and sighting compass) correctly. Not all field staff may have the
skills to undertake distance sampling: minimum standards (i.e. performance criteria) for
observers may need to be instituted.
4. Data collection
Every attempt must be made to collect and record data as accurately as possible: failure to
do so will reduce the accuracy and precision of the density estimate, and hence the value of
the work. Checking of data should be conducted by the field team leader after the first
transects have been sampled (to ensure that data is recorded correctly) and again at the
end of the field trip: any apparent errors/anomalies in the data collection can then be
resolved with the person who collected the data.
5. Data entry and storage
When entering and archiving data, it is useful to think of what information someone else
would need to recreate your analyses in the future if you were not available. Hence, the field
sheets should be scanned and saved as *.pdf documents, and electronic and hard copies
stored and archived. The data should be entered into a spreadsheet or database and
archived, and imported into the latest version of DISTANCE (Thomas et al. 2009) for
analyses. The person entering the data may detect errors/anomalies in the data collection,
and these should be resolved with the field team leader and/or the person who collected the
data.
6. Analysis
The analysis of distance data is a large and growing literature, and general
recommendations are provided in Buckland et al. (2001) and Thomas et al. (2009, 2010).
The analyst(s) should also familiarise themselves with how other macropod distance
sampling studies have been analysed (Coulson & Raines 1985; Southwell 1994; Clancy et
al. 1997). Analyses usually follow three stages (Thomas et al. 2010). First, exploratory data
analysis is conducted to understand the data and identify problems. Particular problems in
wallaby surveys would be too few detections close to the transect line (most usually caused
by animals moving away before the observer detects them) or, as occurred in our study, a
‘spike’ of observations near the line (usually caused by observers rounding small angles
down to zero). It may be useful to assess the data collected by each observer: if an
observer’s data are very different from those of competent observers then it may be
appropriate to exclude data from that observer from analysis.
26
The second stage of analysis is model selection. Truncation (left and/or right) of data and
definition of intervals/cutpoints (for histograms, goodness-of-fit and other diagnostic tests)
are conducted at this stage. The detection function is modelled, with four combinations
usually tried (Thomas et al. 2010): uniform key with cosine adjustments, half-normal key with
cosine adjustments, half-normal key with Hermite polynomial adjustments, and hazard-rate
key with simple polynomial adjustments. Histograms, diagnostic tests and AIC are used to
help assess the usefulness of each model, but the primary objective here should be to
obtain defensible density estimates that are fit for your purpose, and at all times the analyst
should be thinking about the likely behaviour and ecology of wallabies in the study area in
relation to the way the line transect sampling was conducted.
The third and final stage of analysis is inference from the best model(s). Model-averaging
and bootstrapping may be appropriate if several models have similar support. Details of
encounter rates and detection probabilities (with associated sampling variances), and plots
of calculated detection functions should be reported along with estimated densities and
abundances for the study area (or strata).
7. Reporting
The final report should include sufficient information so that your study could be duplicated
by others without your assistance. It would therefore be useful to report the electronic
locations of the scanned field sheets, stored data and DISTANCE project. The report should
also include the study objectives and design, the observers used and any training given, and
a detailed summary of the three stages of analysis (i.e. exploratory data analysis, model
selection and inference from the best model(s)).
References and further reading
Borchers, D.L.; Buckland, S.T.; Zucchini, W. 2002: Estimating animal abundance: closed populations.
Springer-Verlag, London. 314 p.
Buckland, S.T.; Anderson, D.R.; Burnham, K.P.; Laake, J.L.; Borchers, D.L.; Thomas, L. 2001:
Introduction to distance sampling: estimating abundance of biological populations. Oxford
University Press, Oxford. 432 p.
Buckland, S.T.; Anderson, D.R.; Burnham, K.P.; Laake, J.L.; Borchers, D.L.; Thomas, L. 2004:
Advanced distance sampling: estimating abundance of biological populations. Oxford University
Press, Oxford. 416 p.
Burnham K.P.; Anderson D.R. 2002: Model selection and multimodel inference: a practical informationtheoretic approach. 2nd edition. Springer-Verlag, New York. 488 p.
Clancy, T.F.; Pople, A.R.; Gibson, L.A. 1997: Comparison of helicopter line transects with walked line
transects for estimating densities of kangaroos. Wildlife Research 24: 397–409.
27
Coulson, G.M.; Raines, J.A. 1985: Methods for small-scale surveys of grey kangaroo populations.
Australian Wildlife Research 12: 119–125.
Southwell, C. 1994: Evaluation of walked line transect counts for estimating macropod density. Journal
of Wildlife Management 45: 650–657.
Thomas, L.; Buckland, S.T.; Rexstad, E.A.; Laake, J.L.; Strindberg, S.; Hedley, S.L.; Bishop, J.R.B.;
Marques, T.A.; Burnham, K.P. 2010: Distance software: design and analysis of distance
sampling surveys for estimating population size. Journal of Applied Ecology 47: 5–14.
Thomas, L.; Laake, J.L.; Rexstad, E.A.; Strindberg, S.; Marques, F.F.C.; Buckland, S.T.; Borchers, D.L.;
Anderson, D.R.; Burnham, K.P.; Burt, M.L.; Hedley, S.L.; Pollard, J.H.; Bishop, J.R.B.; Marques,
T.A. 2009: DISTANCE 6.0. Release 2. Research Unit for Wildlife Population Assessment,
University of St. Andrews, UK. http://www.ruwpa.st-and.ac.uk/distance/
Thompson; W.L.; White, G.C.; Gowan, C. 1998. Monitoring vertebrate populations. Academic Press,
Inc., San Diego. 365 p.
Warburton, B. 2005: Bennett’s wallaby. In King, C.M. (Ed.): The handbook of New Zealand mammals.
2nd edition. Oxford University Press, Melbourne.
Warburton, B.; Frampton, C. 1993: Monitoring Bennett’s wallaby in South Canterbury. Landcare
Research Contract Report (LC9293/112), Manaaki Whenua—Landcare Research, Lincoln.
31 p.
Westbrooke, I.M.; Etheridge, N.D.; Powlesland, R.G. 2003: Comparing methods for assessing mortality
impacts of an aerial 1080 pest control operation on tomtits (Petroica macrocephala toitoi) in
Tongariro Forest. New Zealand Journal of Ecology 27: 115–123.
Appendix A
The following Department of Conservation documents are referred to in this method:
docdm-146272
Standard inventory and monitoring project plan
28

Wallaby counts - Department of Conservation

Transcription

Similar documents

Wallaby Cambridge English

2011 Bil Jac Sponsorship Flyer

Wallaby Phototherapy Set-up

AGENCY Arc Worldwide Manila, Philippines Creative

Official Zoning Map - Town of Carlton Landing

Intoximeters, Inc.Alco-Sensor FST

Biodiversity synthesis report 2010

ARBOREAL MAMMAL SPOTLIGHT SURVEY 2014

Vertical distribution and population structure of Copepoda in a water

Protocole CIGESMED : Coralligenous based Indicators to Evaluate

Declared animal pests Declared animal pests

Fundamentals of Statistics